Improved Appliance Transient Feature Extraction Via Template Matching

Abstract

In event-based non-intrusive load monitoring (NILM), appliance transient features are usually extracted by event detection and used to identify their operating states. However, the long transients of appliances may be detected incompletely by the existing event detection methods based on general prior knowledge, which leads to the inconsistency of the extracted transient feature samples, and thus impacts the NILM performance. To fill in this gap, assuming that the true starting point of appliance state transition lies in the temporal neighborhood centering the starting point of detected event, an improved appliance transient feature extraction method via appliance-specific template matching is proposed for the event- based NILM approaches. In the proposed method, a variable-length sliding window is adopted to search for the desired transient feature sample, with each point in the temporal neighborhood as its starting point, and its ending point lies in an adaptively defined temporal range based on the appliance-specific transient template information. In addition, the related parameters and thresholds are adaptively updated based on the feature extraction and identification results. For validation, the proposed method is tested on public EMBED and public Pecan Street datasets and two private datasets by comparing with a state-of-the-art event-based NILM benchmark with improved performance.